WITHDRAWN: Genome-wide risk prediction of primary open-angle glaucoma across multiple ancestries.

This manuscript has been withdrawn by medRxiv following a formal request by the QIMR Berghofer Medical Research Institute Research Integrity Office owing to lack of author consent.

Study on Saccharide-Glucose Receptor Interactions with the Use of Surface Plasmon Resonance.

The aim of this study was to investigate the process of attachment of saccharide particles differing in degree of complexity to cell receptors responsible for transport of glucose across the cell membrane (GLUT proteins). This phenomenon is currently considered when designing modern medicines, e.g., peptide drugs to which glucose residues are attached, enabling drugs to cross the barrier of cell membranes and act inside cells. This study aims to help us understand the process of assimilation of polysaccharide nanoparticles by tumour cells. In this study, the interactions between simple saccharides (glucose and sucrose) and dextran nanoparticles with two species of GLUT proteins (GLUT1 and GLUT4) were measured using the surface plasmon resonance technique. We managed to observe the interactions of glucose and sucrose with both applied proteins. The lowest concentration that resulted in the detection of interaction was 4 mM of glucose on GLUT1. Nanoparticles were measured using the same proteins with a detection limit of 40 mM. These results indicate that polysaccharide nanoparticles interact with GLUT proteins. The measured strengths of interactions differ between proteins; thus, this study can suggest which protein is preferable when considering it as a mean of nanoparticle carrier transport.

Influence of GO-Antisense miRNA-21 on the Expression of Selected Cytokines at Glioblastoma Cell Lines.

Introduction: Graphene oxide (GO) is a single layer of carbon atoms with unique properties, which are beneficial due to its surface functionalisation by miRNA. miRNAs are a non-coding small form of RNA that suppress the expression of protein-coding genes by translational repression or degradation of messenger RNA. Antisense miRNA-21 is very promising for future investigation in cancer therapy. This study aimed to detect cytokine expression levels after the administration of GO-antisense miRNA-21 into U87, U118, U251 and T98 glioma cell lines. Methods: U87, U118, U251 and T98 glioma cell line were investigated in term of viability, human cytokine expression level at protein and genes after treatment with GO, GO-antisense miRNA-21 and antisense miRNA-21. The delivery of antisense miRNA-21 into the glioma cell at in vitro investigation were conducted by GO based transfection and electroporation. Results: The results of the protein microarray and gene expression profile showed that complexes of GO-antisense miRNA-21 modified the metallopeptidase inhibitor 2 (TIMP-2), interleukin-6 (IL-6), interleukin 8 (IL-8), intercellular adhesion molecule 1 (ICAM-1), and monocyte chemoattractant protein-1 (MCP-1) expression level compared to transfection by electroporation of antisense miRNA-21 at investigated glioblastoma cell lines. The TIMP-2 protein and gene expression level was upregulated after antisense miRNA-21 delivery by GO complex into U87, U251 and T98 glioblastoma cell lines comparing to the non-treated control group. The downregulation at protein expression level of ICAM - 1 was observed at U87, U118, U251 and T98 glioma cell lines. Moreover, the IL-8 expression level at mRNA for genes and protein was decreased significantly after delivery the antisense-miRNA-21 by GO compared to electroporation as a transfection method. Discussion: This work demonstrated that the graphene oxide complexes with antisense miRNA-21 can effectively modulate the cytokine mRNA and protein expression level at U87, U118, U251 and T98 glioma cell lines.

Exploring Extracellular Vesicles of Probiotic Yeast as Carriers of Biologically Active Molecules Transferred to Human Intestinal Cells.

Extracellular vesicles (EVs) are nanoparticles containing various bioactive cargos-e.g., proteins, RNAs, and lipids-that are released into the environment by all cell types. They are involved in, amongst other functions, intercellular communication. This article presents studies on EVs produced by the probiotic yeast Saccharomyces boulardii CNCM I-745. The size distribution and concentration of EVs in the liquid culture of yeast were estimated. Moreover, the vesicles of S. boulardii were tested for their cytotoxicity against three model human intestinal cell lines. This study did not show any significant negative effect of yeast EVs on these cells under tested conditions. In addition, EVs of S. boulardii were verified for their ability to internalize in vitro with human cells and transfer their cargo. The yeast vesicles were loaded with doxorubicin, an anticancer agent, and added to the cellular cultures. Subsequently, microscopic observations revealed that these EVs transferred the compound to human intestinal cell lines. A cytotoxicity test confirmed the activity of the transferred doxorubicin. Detailed information about the proteins present in EVs might be important in terms of exploring yeast EVs as carriers of active molecules. Thus, proteomic analysis of the EV content was also conducted within the present study, and it allowed the identification of 541 proteins after matching them to the Saccharomyces Genome Database (SGD). Altogether, this study provides strong evidence that the EVs of the probiotic CNCM I-745 strain could be considered a drug delivery system.

Generalising uncertainty improves accuracy and safety of deep learning analytics applied to oncology.

Uncertainty estimation is crucial for understanding the reliability of deep learning (DL) predictions, and critical for deploying DL in the clinic. Differences between training and production datasets can lead to incorrect predictions with underestimated uncertainty. To investigate this pitfall, we benchmarked one pointwise and three approximate Bayesian DL models for predicting cancer of unknown primary, using three RNA-seq datasets with 10,968 samples across 57 cancer types. Our results highlight that simple and scalable Bayesian DL significantly improves the generalisation of uncertainty estimation. Moreover, we designed a prototypical metric-the area between development and production curve (ADP), which evaluates the accuracy loss when deploying models from development to production. Using ADP, we demonstrate that Bayesian DL improves accuracy under data distributional shifts when utilising 'uncertainty thresholding'. In summary, Bayesian DL is a promising approach for generalising uncertainty, improving performance, transparency, and safety of DL models for deployment in the real world.

Portable Surface Plasmon Resonance Detector for COVID-19 Infection.

Methods based on nucleic acid detection are currently the most commonly used technique in COVID-19 diagnostics. Although generally considered adequate, these methods are characterised by quite a long time-to-result and the necessity to prepare the material taken from the examined person-RNA isolation. For this reason, new detection methods are being sought, especially those characterised by the high speed of the analysis process from the moment of sampling to the result. Currently, serological methods of detecting antibodies against the virus in the patient's blood plasma have attracted much attention. Although they are less precise in determining the current infection, such methods shorten the analysis time to several minutes, making it possible to consider them a promising method for screening tests in people with suspected infection. The described study investigated the feasibility of a surface plasmon resonance (SPR)-based detection system for on-site COVID-19 diagnostics. A simple-to-use portable device was proposed for the fast detection of anti-SARS-CoV-2 antibodies in human plasma. SARS-CoV-2-positive and -negative patient blood plasma samples were investigated and compared with the ELISA test. The receptor-binding domain (RBD) of spike protein from SARS-CoV-2 was selected as a binding molecule for the study. Then, the process of antibody detection using this peptide was examined under laboratory conditions on a commercially available SPR device. The portable device was prepared and tested on plasma samples from humans. The results were compared with those obtained in the same patients using the reference diagnostic method. The detection system is effective in the detection of anti-SARS-CoV-2 with the detection limit of 40 ng/mL. It was shown that it is a portable device that can correctly examine human plasma samples within a 10 min timeframe.

Evaluation of Mechanical Properties of 3D-Printed Polymeric Materials for Possible Application in Mouthguards.

Custom mouthguards are used in various sports disciplines as a protection for teeth, temporomandibular joints, and soft tissues of the oral cavity from impact forces. The purpose of this research was to evaluate the mechanical properties of flexible polymeric 3D-printable materials and to select a material with the most favourable physical properties for making intraoral protectors. Four 3D-printable polymeric materials were selected for the evaluation: IMPRIMO LC IBT (Scheu-Dental, Iserlohn, Germany), Keyortho IBT (EnvisionTEC, Gladbeck, Germany), IBT (Formlabs, Somerville, MA, USA), and Ortho IBT (NextDent, Utrecht, Netherlands). A total of 176 samples (44 from each material) was 3D-printed using the stereolitography (SLA) technique. Tensile strength, flexural strength, notch-toughness, Shore hardness, sorption, and solubility tests were conducted. The materials were compared using a series of analyses of variance (one-way ANOVA) with Bonferroni post hoc tests. Statistical analyses were performed with the use of IBM SPSS Statistics 28.0.0 software (IBM, New York, NY, USA). Each material was assigned a score from 1 to 4 depending on the individual test results, and tests were given indexes according to the significance of the parameter in the mouthguard protective function. The number of points obtained by each material in each test was then multiplied by the test index, and the results were tabulated. The material with the highest result among the ones studied-most suitable for the application in mouthguard fabrication-was Keyortho IBT from EnvisionTEC.

In vitro immune evaluation of adenoviral vector-based platform for infectious diseases.

New prophylactic vaccine platforms are imperative to combat respiratory infections. The efficacy of T and B memory cell-mediated protection, generated through the adenoviral vector, was tested to assess the effectiveness of the new adenoviral-based platforms for infectious diseases. A combination of adenovirus AdV1 (adjuvant), armed with costimulatory ligands (ICOSL and CD40L), and rRBD (antigen: recombinant nonglycosylated spike protein rRBD) was used to promote the differentiation of T and B lymphocytes. Adenovirus AdV2 (adjuvant), without ligands, in combination with rRBD, served as a control. In vitro T-cell responses to the AdV1+rRBD combination revealed that CD8+ platform-specific T-cells increased (37.2 ± 0.7% vs. 23.1 ± 2.1%), and T-cells acted against SARS-CoV-2 via CD8+TEMRA (50.0 ± 1.3% vs. 36.0 ± 3.2%). Memory B cells were induced after treatment with either AdV1+rRBD (84.1 ± 0.8% vs. 82.3 ± 0.4%) or rRBD (94.6 ± 0.3% vs. 82.3 ± 0.4%). Class-switching from IgM and IgD to isotype IgG following induction with rRBD+Ab was observed. RNA-seq profiling identified gene expression patterns related to T helper cell differentiation that protect against pathogens. The analysis determined signaling pathways controlling the induction of protective immunity, including the MAPK cascade, adipocytokine, cAMP, TNF, and Toll-like receptor signaling pathway. The AdV1+rRBD formulation induced IL-6, IL-8, and TNF. RNA-seq of the VERO E6 cell line showed differences in the apoptosis gene expression stimulated with the platforms vs. mock. In conclusion, AdV1+rRBD effectively generates T and B memory cell-mediated protection, presenting promising results in producing CD8+ platform-specific T cells and isotype-switched IgG memory B cells. The platform induces protective immunity by controlling the Th1, Th2, and Th17 cell differentiation gene expression patterns. Further studies are required to confirm its effectiveness.

Dimensions of Hybrid and Nanohybrid Mouthguards for Mixed Martial Arts Fighters-Evaluation of a New Method of Fabrication.

Ethylene vinyl acetate mouthguards are the most often used custom protective intraoral appliances by combat sports practitioners. However, due to the difficulties in maintaining the hygiene of such mouthguards and thinning during fabrication, resulting in difficulty in predicting final dimensions, they may not be the optimal solution. The aim of this research was to evaluate an innovative method of mouthguard formation using intraoral modeling of the mouthguard pattern, hybrid acrylic material, and the addition of ZnO nanoparticles. Seventeen mouthguards patterns and 34 custom mouthguards were evaluated: 17 hybrid and 17 nanohybrid. A total of 1122 measurements were performed: each mouthguard and pattern was measured at 22 points. Statistical analyses were performed with the use of IBM® SPSS® Statistics 27.0.0 software (IBM, Armonk, NY USA). The mean thickness of the patterns and mouthguards at all labial areas of central incisors were between 4.65 and 4.80 mm. The thickness at the buccal surface of the first molar was between 3.71 and 4 mm, and at the occlusal surface between 3.40 and 3.56 mm in the cusp area. All measurements of hybrid and nanohybrid mouthguards were strongly and highly correlated with the measurements of the mouthguard patterns. Hybrid and nanohybrid mouthguards are an advantageous alternative to thermoformed custom appliances.

Comparative Evaluation of Different Surface Coatings of Fe3O4-Based Magnetic Nano Sorbent for Applications in the Nucleic Acids Extraction.

Nucleic acid extraction and purification are crucial steps in sample preparation for multiple diagnostic procedures. Routine methodologies of DNA isolation require benchtop equipment (e.g., centrifuges) and labor-intensive steps. Magnetic nanoparticles (MNPs) as solid-phase sorbents could simplify this procedure. A wide range of surface coatings employs various molecular interactions between dsDNA and magnetic nano-sorbents. However, a reliable, comparative evaluation of their performance is complex. In this work, selected Fe3O4 modifications, i.e., polyethyleneimine, gold, silica, and graphene derivatives, were comprehensively evaluated for applications in dsDNA extraction. A family of single batch nanoparticles was compared in terms of morphology (STEM), composition (ICP-MS/MS and elemental analysis), surface coating (UV-Vis, TGA, FTIR), and MNP charge (ζ-potential). ICP-MS/MS was also used to unify MNPs concentration allowing a reliable assessment of individual coatings on DNA extraction. Moreover, studies on adsorption medium (monovalent vs. divalent ions) and extraction buffer composition were carried out. As a result, essential relationships between nanoparticle coatings and DNA adsorption efficiencies have been noticed. Fe3O4@PEI MNPs turned out to be the most efficient nano sorbents. The optimized composition of the extraction buffer (medium containing 0.1 mM EDTA) helped avoid problems with Fe3+ stripping, which improved the validity of the spectroscopic determination of DNA recovery.

Protein-Mediated Transformations of Superparamagnetic Nanoparticles Evidenced by Single-Particle Inductively Coupled Plasma Tandem Mass Spectrometry: A Disaggregation Phenomenon.

Progress toward translating superparamagnetic iron oxide nanoparticles (SPIONs) with specific diagnostic and therapeutic properties for clinical applications depends on developing and implementing appropriate methodologies that would allow in-depth characterizations of their behavior in a real biological environment. Herein, we report a versatile approach for studying interactions between SPIONs and proteins using single-particle inductively coupled plasma tandem mass spectrometry. By monitoring the changes in the size distribution upon exposure to human serum, the formation of stable protein corona is revealed, accompanied by particle disaggregation.

Preliminary studies of the impact of food components on nutritional properties of nanoparticles.

Nowadays consumers have constantly exposed to nanoparticles (NPs) ingestion. Although the impact of NPs on the human has been studied by many authors, they did not consider the influence of food matrix components on bioaccessibility of NPs. This fact has encouraged us to investigate the influence of different food components on NPs. The investigation has been carried out to assess the influence of main food components on the MNPs (metallic nanoparticles) fate during the in vitro gastrointestinal simulation. The experiments have been carried out with the single-particle inductively coupled plasma mass spectrometry (SP-ICP-MS) as a tool for quantitative and qualitative analysis and the scanning transmission electron microscopy (STEM) as a means of qualitative analysis. The influence of various food components on NPs has been confirmed and it may be concluded that the matrix has an impact on the size and form of NPs. The presence of food components significantly changes the behaviour of NPs during simulated gastrointestinal digestion. Possible explanations of the influence of main nutrient groups, i.e. lipids, protein, salts, saccharides and vitamins on NPs have been proposed.

The Influence of Decontamination Procedures on the Surface of Two Polymeric Liners Used in Prosthodontics.

Polymeric liners are materials commonly used in prosthodontics to reshape denture surfaces contacting the soft tissues of the oral cavity. The aim of the study was to determine the impact of different cleaning methods on two polymeric materials used in prosthodontics as non-adhesive permanent liners. The material for the research consisted of samples made from Mollosil Plus (Detax, Ettlingen, Germany)-direct polysiloxan liner; and Plastitanium (Pressing Dental, San Marino, Republic of San Marino)-an injection-molded liner. A total of 198 samples were made, 99 of each assessed material. They were exposed to different cleaning methods-a toothbrush, a toothbrush and soap, a toothbrush and toothpaste (BlendaMed, Procter&Gamble, Cincinnati, OH, USA), a toothpaste and denture cleaning paste (Protefix Hygiene Denture Paste, Queisser Pharma, Germany), denture cleansing tablets (Protefix Hygiene Cleaning Tablets, Queisse Pharma, Germany), and a disinfecting spray (Aftermat, Port Jefferson Station, New York City, NY, USA)-for 1 min, 5 min, 10 min, and 15 min. The image acquisition was performed with scanning electron microscopy and samples were analyzed for the homogeneity of their surfaces-the presence of holes, grooves, precipitate, and small and large separating pieces of the material marking departures from this homogeneity. For each type of damage, one point was given. Continuous data from two groups were compared with Mann-Whitney U testing. Due to a small sample size and distribution of variables other than normal, to compare more than two groups, Kruskal-Wallis testing with post hoc analysis (Dunn test with Bonferroni correction) was used. Categorical data were compared with the chi-square test and the Fisher's exact test. The Mollosil Plus material should be decontaminated with the use of a toothbrush or toothbrush with soap, while Plastitanium material should be disinfected. Plastitanium samples are more susceptible to damage during the decontamination procedures than Mollosil Plus.

Physicochemical and Biological Properties of Graphene-Oxide-Coated Metallic Materials.

In this article, we present graphene oxide (produced by a modified Hummers' method) coatings obtained using two different methods: electrophoretic deposition on 316L stainless steel and chemical modification of the surface of gold applied to the steel. The coating properties were characterized by microscopic and spectrometric techniques. The contact angle was also determined, ranging from 50° to 70°. Our results indicated that GO coatings on steel and gold were not toxic towards L929 cells in a direct cell adhesion test-on all tested materials, it was possible to observe the growth of L929 cells during 48 h of culture. The lack of toxic effect on cells was also confirmed in two viability tests, XTT and MTT. For most of the tested materials, the cell viability was above 70%. They showed that the stability of the coating is the crucial factor for such GO coatings, and prove that GO in the form of coating is non-toxic; however, it can show toxicity if detached from the surface. The obtained materials also did not show any hemolytic properties, as the percentage of hemolysis was on the level of the negative control, which is very promising in the light of future potential applications.

MicroRNA Delivery by Graphene-Based Complexes into Glioblastoma Cells.

Glioblastoma (GBM) is the most common primary and aggressive tumour in brain cancer. Novel therapies, despite achievements in chemotherapy, radiation and surgical techniques, are needed to improve the treatment of GBM tumours and extend patients' survival. Gene delivery therapy mostly uses the viral vector, which causes serious adverse events in gene therapy. Graphene-based complexes can reduce the potential side effect of viral carries, with high efficiency of microRNA (miRNA) or antisense miRNA delivery to GBM cells. The objective of this study was to use graphene-based complexes to induce deregulation of miRNA level in GBM cancer cells and to regulate the selected gene expression involved in apoptosis. The complexes were characterised by Fourier transform infrared spectroscopy (FTIR), scanning transmission electron microscopy and zeta potential. The efficiency of miRNA delivery to the cancer cells was analysed by flow cytometry. The effect of the anticancer activity of graphene-based complexes functionalised by the miRNA sequence was analysed using 2,3-bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxyanilide salt (XTT) assays at the gene expression level. The results partly explain the mechanisms of miRNA deregulation stress, which is affected by graphene-based complexes together with the forced transport of mimic miR-124, miR-137 and antisense miR-21, -221 and -222 as an anticancer supportive therapy.

Simple Ultraviolet-Visible Spectroscopy-Based Assay for Fast Evaluation of Magnetic Nanoparticle Selectivity Changes After Doping.

Simple ultraviolet-visible spectroscopy-based methodology was proposed and utilized for the initial characterization of potential changes in selectivity of doped magnetic nanoparticles. Doped and undoped iron(II,III) (Fe3O4) magnetic nanoparticles were synthesized by the coprecipitation method. The doping processes of nanoparticles were confirmed using optical emission spectrometry, while the sizes of undoped and Cu-doped nanoparticles were investigated using a high-resolution field emission scanning electron microscope. The average diameters of nanoparticles were 8.34±1.78 nm and 9.12±1.93 nm, for doped and undoped materials, respectively. The influence of the nanoparticle's doping on their selectivity towards chosen analyte was monitored by the spectral techniques such as ultraviolet-visible and optical emission spectrometry. The interaction between Cu-doped Fe3O4 nanoparticles and cuprizone (a compound forming the characteristic colorful complex with copper) was confirmed. The elaborated studies proved the potential of ultraviolet-visible spectroscopy for the fast qualification of magnetic nanoparticles in terms of their ability to separate the selected analyte from the sample matrix.

Automated AI labeling of optic nerve head enables insights into cross-ancestry glaucoma risk and genetic discovery in >280,000 images from UKB and CLSA.

Cupping of the optic nerve head, a highly heritable trait, is a hallmark of glaucomatous optic neuropathy. Two key parameters are vertical cup-to-disc ratio (VCDR) and vertical disc diameter (VDD). However, manual assessment often suffers from poor accuracy and is time intensive. Here, we show convolutional neural network models can accurately estimate VCDR and VDD for 282,100 images from both UK Biobank and an independent study (Canadian Longitudinal Study on Aging), enabling cross-ancestry epidemiological studies and new genetic discovery for these optic nerve head parameters. Using the AI approach, we perform a systematic comparison of the distribution of VCDR and VDD and compare these with intraocular pressure and glaucoma diagnoses across various genetically determined ancestries, which provides an explanation for the high rates of normal tension glaucoma in East Asia. We then used the large number of AI gradings to conduct a more powerful genome-wide association study (GWAS) of optic nerve head parameters. Using the AI-based gradings increased estimates of heritability by ∼50% for VCDR and VDD. Our GWAS identified more than 200 loci associated with both VCDR and VDD (double the number of loci from previous studies) and uncovered dozens of biological pathways; many of the loci we discovered also confer risk for glaucoma.

Analysis of common genetic variation and rare CNVs in the Australian Autism Biobank.

BACKGROUND: Autism spectrum disorder (ASD) is a complex neurodevelopmental condition whose biological basis is yet to be elucidated. The Australian Autism Biobank (AAB) is an initiative of the Cooperative Research Centre for Living with Autism (Autism CRC) to establish an Australian resource of biospecimens, phenotypes and genomic data for research on autism. METHODS: Genome-wide single-nucleotide polymorphism genotypes were available for 2,477 individuals (after quality control) from 546 families (436 complete), including 886 participants aged 2 to 17 years with diagnosed (n = 871) or suspected (n = 15) ASD, 218 siblings without ASD, 1,256 parents, and 117 unrelated children without an ASD diagnosis. The genetic data were used to confirm familial relationships and assign ancestry, which was majority European (n = 1,964 European individuals). We generated polygenic scores (PGS) for ASD, IQ, chronotype and height in the subset of Europeans, and in 3,490 unrelated ancestry-matched participants from the UK Biobank. We tested for group differences for each PGS, and performed prediction analyses for related phenotypes in the AAB. We called copy-number variants (CNVs) in all participants, and intersected these with high-confidence ASD- and intellectual disability (ID)-associated CNVs and genes from the public domain. RESULTS: The ASD (p = 6.1e-13), sibling (p = 4.9e-3) and unrelated (p = 3.0e-3) groups had significantly higher ASD PGS than UK Biobank controls, whereas this was not the case for height-a control trait. The IQ PGS was a significant predictor of measured IQ in undiagnosed children (r = 0.24, p = 2.1e-3) and parents (r = 0.17, p = 8.0e-7; 4.0% of variance), but not the ASD group. Chronotype PGS predicted sleep disturbances within the ASD group (r = 0.13, p = 1.9e-3; 1.3% of variance). In the CNV analysis, we identified 13 individuals with CNVs overlapping ASD/ID-associated CNVs, and 12 with CNVs overlapping ASD/ID/developmental delay-associated genes identified on the basis of de novo variants. LIMITATIONS: This dataset is modest in size, and the publicly-available genome-wide-association-study (GWAS) summary statistics used to calculate PGS for ASD and other traits are relatively underpowered. CONCLUSIONS: We report on common genetic variation and rare CNVs within the AAB. Prediction analyses using currently available GWAS summary statistics are largely consistent with expected relationships based on published studies. As the size of publicly-available GWAS summary statistics grows, the phenotypic depth of the AAB dataset will provide many opportunities for analyses of autism profiles and co-occurring conditions, including when integrated with other omics datasets generated from AAB biospecimens (blood, urine, stool, hair).

Verifying explainability of a deep learning tissue classifier trained on RNA-seq data.

For complex machine learning (ML) algorithms to gain widespread acceptance in decision making, we must be able to identify the features driving the predictions. Explainability models allow transparency of ML algorithms, however their reliability within high-dimensional data is unclear. To test the reliability of the explainability model SHapley Additive exPlanations (SHAP), we developed a convolutional neural network to predict tissue classification from Genotype-Tissue Expression (GTEx) RNA-seq data representing 16,651 samples from 47 tissues. Our classifier achieved an average F1 score of 96.1% on held-out GTEx samples. Using SHAP values, we identified the 2423 most discriminatory genes, of which 98.6% were also identified by differential expression analysis across all tissues. The SHAP genes reflected expected biological processes involved in tissue differentiation and function. Moreover, SHAP genes clustered tissue types with superior performance when compared to all genes, genes detected by differential expression analysis, or random genes. We demonstrate the utility and reliability of SHAP to explain a deep learning model and highlight the strengths of applying ML to transcriptome data.

Genome-wide association study of dietary intake in the UK biobank study and its associations with schizophrenia and other traits.

Motivated by observational studies that report associations between schizophrenia and traits, such as poor diet, increased body mass index and metabolic disease, we investigated the genetic contribution to dietary intake in a sample of 335,576 individuals from the UK Biobank study. A principal component analysis applied to diet question item responses generated two components: Diet Component 1 (DC1) represented a meat-related diet and Diet Component 2 (DC2) a fish and plant-related diet. Genome-wide association analysis identified 29 independent single-nucleotide polymorphisms (SNPs) associated with DC1 and 63 SNPs with DC2. Estimated from over 35,000 3rd-degree relative pairs that are unlikely to share close family environments, heritabilities for both DC1 and DC2 were 0.16 (standard error (s.e.) = 0.05). SNP-based heritability was 0.06 (s.e. = 0.003) for DC1 and 0.08 (s.e = 0.004) for DC2. We estimated significant genetic correlations between both DCs and schizophrenia, and several other traits. Mendelian randomisation analyses indicated a negative uni-directional relationship between liability to schizophrenia and tendency towards selecting a meat-based diet (which could be direct or via unidentified correlated variables), but a bi-directional relationship between liability to schizophrenia and tendency towards selecting a fish and plant-based diet consistent with genetic pleiotropy.